FreeBSD SMPng Interview with Scott Long

animus9 writes "There's an interesting interview with Scott Long over at the ONLamp.com. Scott explains the difference between the various locking methods, and the current status of SMP in FreeBSD 5. He also compares the new SMP implementation with that of FreeBSD 4.x, NetBSD, DragonFly, and Linux. Other items touched upon include scalability, the status of KSE & ULE, and much more."

Sleep locking, spin locking

[Julian Morrison]

This article discusses FreeBSD's preference for sleep locks as versus the spin locks in Linux.

Anybody know why Linux went for the spin lock approach? What are the relative merits?

Re:BSD is starting to look as a viable alternative

[Anonymous Coward]

Hilarious read since I can clearly relate to it. I now see the light having to deal with continual issues with a major network vendor's Linux-based IDS. Over a course of about a year I've deal with issues such as telnet/ssh/https management interface hanging and the only work-around was to bounce the appliance and, separately, disable swap memory with "swapoff -a". Vendor couldn't fix it so they replaced it with a higher end model (same Linux OS). Then, it was the spontaneous reboots with the new model. Sent config and "show tech-support" but nothing obvious was wrong according to vendor. They ended up RMA'ing the IDS. So, I get the replacement, patched it up and applied the latest signature. It runs fine for a week, a new signature update comes out, I apply it but instead of taking the normal ~10 minutes it doesn't complete after several hours later. I opened another case to report the problem but knowing it's not going to be fixed and I can no longer manage it all after a bounce I went ahead and reimage the IDS, go through patching and updating the latest signature again and this time it took it. I now cringe anytime I hear/see anything Linux on our work network. But my eyes lit up the other week when one of the guys in another department demoed Niksun NetVCR. Doing a little prior research I knew it was FreeBSD based and asked him what he thought about stability. He said it was amazing and a huge improvement over Windows XP based Sniffer Distributed (usability wise also). At home, I was fortunate to start out with BSD and since it's had a perfect record with uptime so far I've stuck with BSD for everything I can. Servers (web, DB, sshd, sftp, ntpd, etc.) are either NetBSD or FreeBSD. For firewall/VPN/traffic shaper I use www.m0n0.ch/wall (FreeBSD based). So, talking with one of the unix guys he expressed unhappiness that management is thinking about replacing Sun with Linux. Thinking about the IDS and how it was originally SunOS until the vendor cheapened up and replaced it with Linux I was thinking (probably too was the unix guy), GOD HELP US! I guess it'll be good for job security because if you're not occupied fixing stability issues you'll be patching security holes every other day.

Re:SMP BSD

[setagllib]

OpenBSD basically needs a bigger contributing (at least PRing) user base. So does NetBSD of course. The number of developers is fine in both because they are talented and can work together (oppose thousands of Linux devs who still can't engineer for stability), but the user bases are too small to really test everything at once.

Shame isn't it? With Linux stealing the spotlight all the other more deserving projects are left lacking users.

Re:meh

[setagllib]

The first time I booted 2.6.3 (mainline, under Slackware, previously 2.6.2 with less problems) every file opened for writing would be truncated and never written to. My entire home dir was mangled and it took a very long painful sitting of deleting and re-configuring before things worked again, and even then there were flake outs. I never booted 2.6.3 again.

2.6 then had some large changes (nptl, new SCSI subsystem without warning, etc.) and now at 2.6.10 seems to be at least sort-of stable, but there are compile warnings in wireless drivers I think are actual mistakes and am glad I don't have that hardware. On the other hand, and this is a big thing in favor of the BSD style kernel configuration, with 2.6.10 a certain magical combination of kernel options left me without a console outside of X (and no, I did not ask it to use serial console), and reconfiguring from scratch was the only thing that fixed it. I still have not been able to reproduce this (but remember it happening with an earlier kernel, might have been a 2.4) or even figure out what option/combination caused it. I mean, that's pretty f#ed up.

Re:meh

[setagllib]

Bzzzt try again, troll. NPTL has been in Linux since the middle of the 2.5 development cycle.

/usr/portage/dev-libs/glibc/glibc-2.3.4.20040619-r 2.ebuild
# Minimum kernel version we support
# (Recent snapshots fails with 2.6.5 and earlier)
MIN_KERNEL_VERSION="2.6.5"

Headshot. If it hadn't changed, there'd be no reason that newer user-land nptl libraries wouldn't work with older kernels. Read up before you think you're fighting 'trolls'.

And there was no "new SCSI subsystem" in 2.5 or 2.6 period.

http://www.webservertalk.com/message841936.html [webservertalk.com]
Sorry, really bad wording on my part, based on some confusing Slash comments I read before. Hardly trolling, you'll notice.

Linux does not cater for incompetant people

No, I actually did check everything, and have been configuring and compiling Linux kernels with mostly success (except weird shit like this) for years. There's no magic to it, don't pretend to be a technical expect because you've never found a bug. Same goes for that "absolutely no idea when it comes to kernel coding" assumption: I am a coder and I do know when a warning is an error in disguise. By the looks of it the calling parameters of something internal changed (since this did not happen in 2.6.9) but not all drivers were updated, and nobody cared. If this is not the case, they should fix compile warnings: the BSDs do, because warnings left over in 'stable' branches signify lazy/careless developers (i.e. Linux contributors).

Nice AC posting by the way, if you're going to make insulting claims against someone, do it with your own name or risk not being taken seriously. If I wanted to troll, which I don't, I wouldn't do it under my own name. From this perspective we gather that you're the troll and I'm making honest observations. Have a really bad day, you deserve it.